Note updates

Adds the [Lecture 5 Detailed Explanation](/notes/csci678/lecture-5-explanation) and [Full Handout](/notes/csci678/lecture-5-full-translation). Derives tree symmetrization from minimax duality and martingale differences, explaining dependent data, binary relabeling, and contraction conditions. Both documents have matching Chinese editions through the language switcher. The source handout is not published online; all original results, equation numbers, algorithms, and footnote information are preserved, with editorial additions explicitly distinguished.

Adds the [Lecture 6 Detailed Explanation](/notes/csci678/lecture-6-explanation) and [Full Handout](/notes/csci678/lecture-6-full-translation). Develops sequential Sauer, threshold and monotone counterexamples, regression entropy integrals, and surrogate normalization. Both documents have matching Chinese editions through the language switcher. The source handout is not published online; all original results, equation numbers, algorithms, and footnote information are preserved, with editorial additions explicitly distinguished.

Adds the [Lecture 7 Detailed Explanation](/notes/csci678/lecture-7-explanation) and [Full Handout](/notes/csci678/lecture-7-full-translation). Derives the exponential-weights potential, mistake-indexed experts, and the Perceptron bound, with margin and dimension qualifications. Both documents have matching Chinese editions through the language switcher. The source handout is not published online; all original results, equation numbers, algorithms, and footnote information are preserved, with editorial additions explicitly distinguished.

CSCI 678 adds the [Lecture 4 Detailed Explanation](/notes/csci678/lecture-4-explanation) and [Full Handout](/notes/csci678/lecture-4-full-translation). They connect fat-shattering and monotone-class learning rates with neural-network covers, spectral complexity, and margin generalization bounds. The explanation adds the atom-sampling proof, internal-cover constants, the KKT calculation for layerwise error allocation, and a two-layer numerical example. Both original experiment figures are retained, with explicit notes on normalization, endpoints, and the abbreviated high-probability statement. Matching Chinese pages are available through the language switcher.

The CSCI 678 project adds Lecture 1 and 2 materials, two companion documents per lecture: the [Lecture 1 explanation](/notes/csci678/lecture-1-explanation) with its [full slide translation](/notes/csci678/lecture-1-full-translation), and the [Lecture 2 explanation](/notes/csci678/lecture-2-explanation) with its [full slide translation](/notes/csci678/lecture-2-full-translation). The English text of the slide translations is restored from the original slides. Notes for Lectures 1-3 are now all online.

The CSCI 678 project adds the [Lecture 3 Full Translation](/notes/csci678/lecture-3-full-translation): the complete content of the three parts — covering functions, covering projections, Dudley's entropy integral with the chaining technique, and pseudo-dimension — with mathematical notation, theorem numbering, and proof structure preserved from the original slides. Read it alongside the [Lecture 3 Preview Report](/notes/csci678/lecture-3-preview).

Study notes for CSCI 678 (Theoretical Machine Learning, Fall 2026, Haipeng Luo) join this site as a standalone Project. The first document is the [Lecture 3 Preview Report](/notes/csci678/lecture-3-preview), covering pointwise covering, sample-projection covering, Dudley's entropy integral, and pseudo-dimension for infinite real-valued function classes. Notes for later lectures will be added over time.